Numpy is great if you can vectorize your code in a straightforward manner. But when you start doing anything non-standard it becomes difficult. I can tell you

Numpy is great if you can vectorize your code in a straightforward manner. But when you start doing anything non-standard it becomes difficult. I can tell you that programmer productivity can plummet drastically when you need to get everything into vectorized form. Also, custom classes and types do not vectorize well at all.